Boosting Algorithm's Applications In Distance Education Analysis And Image Classification

Machine learning and data mining are two active research directions in recent years. Data mining is a process that is able to discover the trends and patterns hidden in large databases.In this paper, Boosting algorithm is studied and adopted in both content-based image classification for web image search engine and student behavior analysis for distance education. The main idea of boosting is to combine many simple and moderately inaccurate classifiers into a single, highly accurate classifier. Moreover, if the weak classifier depends only on a single feature, the boosting training process, which selects a new weak classifier in each stage, can also be viewed as a feature selection process.In distance education, student behavior analysis is an active research topic in recent years. However, most previous works are conducted manually by emphasizing different aspects, i.e. all the factors are selected, categorized and analyzed manually by investigators. Because Boosting is capable of conducting training and feature selection simultaneously, we adapted it into student behavior analysis based on the training data collected from questionnaires. We not only obtained a prediction model that is possible to predict students' academic successes and assist them to adjust their learning behaviors, but also found the most important factors and their relations which influence either learning or teaching in distance education. More importantly, these findings are of great importance to academic administrators, faculty members, and instructional developers to improve the teaching modes and on-line courseware design.As there are numerous images on the Internet, image classification is an essential and important step that is useful to improve the efficiency and accuracy of the image search engine. In this paper, we investigated how to classify images into cartoon images and realistic images, and then classify realistic images into painting images and photo images. By investigating the difference between cartoon images and realistic images, and the difference between painting images and photo images, we proposed a novel feature, i.e. color variance histogram, to improve the classification accuracy. Taking advantage of the feature selection property of Boosting algorithm, we employed a set of low level features for image classification. Experimental results show that the proposed approach is effective in improving the classification accuracy.